The invention provides a long-short time memory model-based idea recognition method. The method comprises the following steps of obtaining electroencephalogram signal data; Extracting data characteristics of the electroencephalogram signal; Performing classified learning on the extracted data features to complete establishment of a network model; And evaluating the performance of the built networkmodel. Electroencephalogram signal characteristics are extracted through an LSTM network model and then classified through a GB classifier, and performance evaluation of the network model is obtained. Experimental results show that the method provided by the invention can combine the LSTM algorithm in deep learning with the traditional GB classifier, and also provides a new direction for subsequent electroencephalogram classification recognition research under the condition that all electroencephalogram signal samples are successfully classified.